SAIL:
April 11, 2025
Welcome to Sensemaking, AI, and Learning (SAIL). I focus on AI and higher education.
I was at the ASU-GSV conference this past week. This is the big annual monied conference where private equity/VC firms, big tech, startups, and some academics meet to try and court one another. It’s an exhausting conference and one that simultaneously causes despair for the future of education and optimism that we might see systems change.
Hardly surprising, but AI was center stage. I like the framing of Agrawal et al. around innovation in the use of AI with a scale on one end focused on point-based solutions (i.e. solve a problem) and the other end focused on systems change. I saw none of the latter here and remain surprised how the traditional university innovators have done little more than sign a contract with an LLM provider and launched a chat bot. The educational AI innovation will not come from inside as there is no vision.
AI and Education
How do students use AI? Students make extensive use of AI and Anthropic releases a solid analysis focusing on variations in different academic disciplines. They present an overarching framework of interaction patters: direct, collaborative, problem solving, and output creation. Our research from a few years ago confirmed that students tend to rely on, rather than learn from, AI. According to this report, that may be changing.
Announcing: Matter and Space. The last 18 months have been the most intense, in terms of learning, of my young life. We’ve been working on building a new learning operating system that employs the critical capabilities that AI offers in an attempt to re-center and ground learning and sensemaking for humanity in a hopeful future. I’ll share more about this in the future, but for now, this is our vision (scroll down for several interviews).
AI in General
OpenAI has a range of use cases posted. Worth reviewing, though education is sparsely represented (i.e. not at all)
ChatGPT now remembers all your past conversations. This may be problematic for memory service companies and people who don’t want full memory if they use ChatGPT across various life roles: i.e. parent and employee. To the first point: it’s hard for an AI startup to nail the right value add. We’ve been focusing on: if AI getting better makes our product better, we have the right focus. If AI getting better threatens us, we need to rethink our value add.
Stanford releases their annual AI report. It weighs in at 400+ pages (clearly a task for Gemini to summarize). Chapter 7 focuses on education, the mainly computer science in both K-12 and higher education. IEEE offers twelve charts detailing the report (with simple commentary.
If you’re tracking voice agents, this is an excellent resource.
Google launches their Agent2Agent protocol: “a shared vision of a future when AI agents, regardless of their underlying technologies, can seamlessly collaborate to automate complex enterprise workflows and drive unprecedented levels of efficiency and innovation”. Some overlap with MCP (which focused on APIs and tools, not agent to agent interaction). Best explainer I’ve seen on MCP is here.
Google launched a range of new AI tools and technologies. Details here. They are slowly becoming the AI leader they should be - they fumbled hard over the last two years.
Shopify’s CEO dropped an internal memo that has been gaining traction. Short version: start with AI before you start with humans.
Meta dropped Llama 4 and will follow up with a few models (including Behemoth that clocks in at 2T parameters). Concerns have been raised about both political leanings (though they detail this in the link above “strong political lean at a rate comparable to Grok”) and gaming LMS benchmarks.
OpenAI is excellent at producing startups by disaffected former employees (Anthropic, SSI). The latest is the former CTO and others who are seeking to raised $2b at a $10b valuation.
Amazon keeps rolling out their AI infrastructure and offerings. The latest: Nova Sonic - human-like voice conversations.
Europe is sharpening their AI focus to better compete with USA and China.
Stop managing AI projects like traditional software. Test and experiment with your AI models and tools to success - rather than traditional SWE roadmap approaches (though those traditional approaches are not completely irrelevant - it’s more about adding better testing questions to guide the roadmap).